Decoding Human Languages: A Transfer Learning Approach for Language Translation for Low Resources Languages - Nepali, Urdu, Pashto and Punjabi
Pankaj Kumar Srimal, Kiran Kumar Makam · 2024
Signal intelligence across the world face the complex task of intercepting cross-border communications due to multiple language and multiple dialect. In natural language processing (NLP) research, text data is more abundant and easier to handle and this abundance makes text a primary focus in NLP research. Speech data, on the other hand, is less common and more difficult to work with due to its expressive features like tone and pitch. As a result, advancements in speech-to-text translation are limited to a few widely spoken languages, creating a void for low resource languages. The paper represents a novel hybrid approach based on transfer learning to enhance speech-to-text translation by de-noising background and accurately interpreting speech into the target text language. The solution has showcased an accuracy of 0.6 and can help signal intelligence analysts identify potential threats, aiding commanders in making informed decisions which will assist in gaining a decisive advantage.